LIBRATUS

No Gravatar

REPRINT OF MY ARTICLE FOR BLUFF EUROPE: JUNE 2017

The End of the World as we Know it

Back in January, the poker world was stunned when four top online pros were humbled by the poker AI program Libratus. For years, people said that no limit Hold’em would never be conquered by AI. They argued that the game was so complex that no computer would be able to figure it out, that it needed human experience and intuition to beat the best players.

Learning from “The Beast”

I learned a long time ago (in the mid-nineties) not to doubt the power of computers. Back in the early 1990’s, I was a keen Chess player and master strength. I stated to anyone that would listen that no program would ever beat a World Champion in proper tournament time limits. By 1997, I was eating my words when Kasparov lost to Deep Blue.

Since then, Chess programs have gone from strength to strength. So much so that Magnus Carlsen (the highest rated player in history) now comes some way behind the top programs. AI has already conquered the best at games like Checkers and Go and Limit hold’em was solved several years ago.

The Last Great Frontier

But in all these games, all of the available information is already known everybody cried! Well I not only have far more faith in AI than I did in the early 1990’s but I also have far more faith in human ingenuity too. The problem of uncertainty in no limit hold’em has been approached and partly solved and with stunning effect!

SNIPER uses FERN/WOM and mimics LIBRATUS’ style in many lines.

Four top online pros were “schooled” in January by an absolute “monster” of a player. Libratus was so strong heads up that one of the players thought that he was being cheated. This wasn’t an accusation, just a statement regarding how good his opponent was. So does this all mean that the end will be nigh for poker….no chance!

For years AI has had a growing influence on our life. If Chess is a great example then we have nothing to fear but fear itself! Sure there were cheats in chess using sophisticated programs to become “super players”. What happened in Chess was that programs became so good that they artificially enhanced the playing standards of the players that were using them.

In the old days, the Chess world was dominated by Russians. While Russia is still at the forefront of World Chess, only three out of the current world top ten are either Russian or from ex-Russian states and only one of the top five. The current World Champion is Norwegian and this can partly be attributed to AI.

FERN/WOM numbers take away many situations where human are apt to go wrong.

AI has quantum leaped the strength of Chess masters and Grandmasters with the number of Super Grandmasters now at an all-time high. Players can now have the best chess “coaches” in the world inside their own home. Bobby Fischer was regarded as the best player to ever play the game for years and had an ELO of 2785 when he became World Champion in 1972, a record that stood for ages. There are currently nine players higher than that total in the world rankings.

AI isn’t something to be feared unless online poker sites don’t prevent people from using it to defeat other online opponents. AI is something that will catapult us towards being better poker players and far better than our ordinary opponents in live games.

Lessons from Chess

While Chess is a game with perfect information, the top human brains behind the advancement of AI were never going to leave poker alone and Libratus was the end product. I have examined the Libratus “Man vs Machine” match in some depth and I believe that the program exposes flaws in human thinking in key situations and recognises patterns that to us are unrecognisable.

The strength of Libratus certainly doesn’t bother me in the same way that the strength of Chess programs never bothered me. Libratus defeated each top online pro heads up and that is much easier to do than in a ring game. Secondly I am not a heads up player and I have started playing more live poker in the past couple of years.

I have been looking at Libratus to see how I can possibly expand my own game. Its thinking process was described as unconventional by all of the human opponents’ that it played back in January.

A match that was played over 20 days and 120,000 hands! The hands were also reversed. This meant that each opponent had the same hands. This makes the result highly statistically significant seeing as the program clipped the pros’ for $1.7m.

Chess players learn from AI to become better players. Financial traders did the same thing years ago and started using algorithms to trade on the financial markets. So the future isn’t bleak. Human beings will constantly over-estimate their own ability. They will constantly look to take shortcuts in their education and through a lack of patience and/or greed, will simply jump into poker with insufficient knowledge.

  • “They will be meat and drink to the elite few who have not only elevated their games, but have also taken the best bits from AI and incorporated it into their own games. The strongest “chess players” in the world aren’t in fact computer programs but program/human hybrid teams that play in man/machine team events. So with two man chess teams, the best team isn’t best human/best human or even best program/best program but best program/best human”

A sure sign that as poker players, we need to be working with AI rather than against it! The technology is here to stay whether we like it or not. We either live in fear or we embrace it and become better poker players to beat up our opponents even more.

Libratus used numerous unconventional lines that could have big long term implications on poker theory. As it stands, these were only heads up matches. So we need to “watch this space” to see if Libratus or its successors can win in the toughest online ring games or against multiple opponents….don’t bet against it!

Carl “The Dean” Sampson is a poker coach and full time pro and you can see him on his blog at www.pokersharkpool.co.uk

ADDENDUM:

  • After this article was written for Bluff Europe magazine, I continued my research on Libratus. In particular on how its style compared with my very own FERN/WOM SNIPER. It’s clear to me now why FERN/WOM and SNIPER are so successful. In fact when I reverse engineered these concepts to attack mid-stakes games, they proved to be absolutely CRUSHING.

Libratus was taught the rules of poker and then learned the strategy for itself. But what EXACTLY is THE STRATEGY? The very nature of strategy itself lies in how humans interact with the game of poker. It is why no one strategy can possibly fit all.

Libratus worked this out to get closer to the true REALITY of the sum of all knowledge….namely to become a pseudo super-user (a term used to describe a poker player who can see their opponents’ cards).

Libratus literally re-wrote poker theory. Its use of ultra-aggressive lines created huge amounts of dead money and fold equity. Over betting the pot was its signature move and it did this move more than the best humans do. This strategy crushed against world class players but WHY? There is only one explanation, Libratus spotted patterns that were impossible to recognise in the heat of battle with the limitations of a human brain. It simply exploited human weakness and a large part of that weakness is in how humans value money….computers do not.

Fundamentally, this is why SNIPER and FERN/WOM are so successful. It shares many of the attributes of Libratus. Some were accidental but many others were intentional and added over the past few months.

SNIPER uses Bayes Theorem at its very core; it just underlines what a POWERFUL OPPONENT it really is for ANY OPPONENT that you could ever face.

Carl “The Dean” Sampson – August 2017

Share